Discussion ability simulated in 1989–1991, but none in the

247 G.M. Richter European Journal of Agronomy 11 1999 239–253 Fig. 5. Relationship between simulated yields Y and plant available water in the A-horizon during years with , 1988 Fig. 4. Relationships between winter rye grain yields tha and and without spring drought , 1989. plant available water mm simulated for all ecotopes n =393 in the catchment during 5 years; example of scatter for 1988 regression analysis, N r explained 8 and 25 in and 1989 ; regression lines for the respective years 1990 and 1991, in addition to water only. The include r 2 values. small influence of potentially mineralizable N on simulated yields is plausible because mineral N linear relationship being distinctly different was not limiting in the management scenario. between years. PAW explained 60–75 of the The yields recorded for the period 1988–1991 modelled yield variability in years with water were analysed by regression for the effects of PAW stress, but very little 10 in 1988 when there and irrigation. No dependency on average field was a good water supply. In the PAW range 46– PAW was found for either the complete or the 146 mm, the yield increased from 5 kg DMmm of annual data sets [Fig. 6a]. PAW had no influence PAW in a wet year 1988 to 39 kg DMmm in a even in drier years as modelling had suggested. No dry year 1989. The simulated yield increase overall effect of applied irrigation water on yield in kg DMmm PAW was inversely related to rain- was detected [Fig. 6b; r 2=0.077]. In individual fall during the period tillering to anthesis years, irrigation significantly increased the yield in RAIN EC61 ; r 2=0.94; p0.01. The interaction 1988 only. Although this year was generally wet, of simulated yield and available soil water in the the late spring was comparatively dry Table 1. Ap horizon was similar Fig. 5, but less clearly Multiple regression analysis for the interaction expressed than for the whole profile. For example, between PAW, irrigation and observed yields sug- in the dry year 1989, the r 2 for the relationship gested a negative effect of irrigation on yields for decreased greatly 0.45 vs. 0.75. This suggests high PAW soils. This is plausible, because water that yields are limited by rooting depth, higher logging on loamy sand increases the probability of yields being obtained where roots can extract water disease. Irrigation may also decrease plant growth from a greater thickness of soil, especially during by leaching N. However, in the simulation, irriga- later periods of plant growth EC 31–87. tion enhanced the N-uptake and -export with the The effect of N supplied by mineralization of harvest increasing nitrate leaching by only 8– soil organic matter and residue N r , kgha on the 12 kghayear. variability of grain yield was about an order of magnitude less than that of PAW. As a single component, it accounted for 29–51 of the vari-

4. Discussion ability simulated in 1989–1991, but none in the

year of 1987–1988. This is possibly an artefact caused by its inherent relation to soil texture and The results are significant for scaling-up meth- odology and model transfer to other species. water-holding capacity. According to multiple 248 G.M. Richter European Journal of Agronomy 11 1999 239–253 tion in the ratio of observed to simulated yields. Currently discussed scaling approaches Downing, 1997; Russell and van Gardingen, 1997 propose distributed inputs to describe lower regional yields, but it was shown here that the model still overesti- mates the catchment mean. Simulations could use aggregated inputs Table 6, but additional tempo- ral scaling was necessary for the means Table 5 and distribution of yields in the catchment Fig. 7. The temporal variability of weather was found to be more important than the spatial variability of the soil. Both issues need further discussion with respect to data quality and model sensitivity. Details for the soil input used in the simulation depend on the a priori knowledge of landscape complexity. The particular catchment in northwest Germany was located on the edge of heathlands and was atypically complex. Although it was domi- nated by podsols, parts were covered by fine a b Fig. 6. Dependence of observed winter rye grain yield Y, tha on a plant available water and b applied irrigation water; slope of regression line in 1988 DYDI; tDMha per mm water. Hypothesis A, that detail of soil input data may be reduced without losing information critical to the mean representation by the model, was con- firmed for all years and the overall mean of this particular catchment. However, hypothesis B, that the output may be scaled by a regionally unique Fig. 7. Distribution of time-scaled winter rye yields tha simu- scaling factor de Koning and van Diepen, 1993, lated across the catchment for all years 1988–1991 and single year 1990. could not be confirmed due to considerable varia- Table 6 Comparison of plant available water and simulated winter rye grain yield using soil survey data at different resolution; local map 1:5000 vs. EU Soil Map 1:1 000 000 +pedotransfer function Groenendijk, 1989; scaled mean includes standard deviation in parentheses Soil Series Area weight PAW mm Yield tha 1:5000 1:10 6 1:5000 1:10 6 Podsol 60 63 50 6.3 5.7 Podsoluvisol 40 107 129 7.5 7.2 Weighted mean 6.8 6.4 Scaled mean 5.2 0.9 5.1 0.7 249 G.M. Richter European Journal of Agronomy 11 1999 239–253 textured colluvial soils leading to very variable and 2 larger harvest losses under field conditions compared to experimental plots. This would justify drainage and nitrate-leaching rates Richter et al., 1998, which were bimodally distributed. However, using even a temporally variable scaling factor, YOBSYMOD, which overall was similar to that the bimodal distribution of PAW had no effect on the overall distribution of scaled or unscaled yields reported for wheat by de Koning and van Diepen 1993. The results from local, non-irrigated State in the catchment at any time Fig. 7. Differences in the unscaled mean yields of each soil type were Variety Trials were close to obtainable yields and thus support the scaling of yields for technological not significant, and scaling with a mean annual CF had eliminated some of the individual distribu- reasons. The detected negative influence of increas- ing PAW on recorded yield can be attributed to tion patterns of unscaled yields. However, there was no relation between the field specific annual adverse environmental effects and unfavourable microclimate leading to diseases, which were not CF and soil properties, and a mean annual CF seemed justified. More importantly, the hydrologi- included in the model and would have been pre- vented in experimental plots. cal sub-model using capacity parameters is rela- tively robust, and plants are able to extract water In two of the years, the simulated yield was wrongly affected by environmental variables: in through the profile, effectively integrating PAW. The approximate equality of the mean of the 1990 by the radiation–temperature regime, and in 1992 by the water deficit. The problems encoun- simulations and the simulation of the mean sug- gests a linear relationship between parameters and tered in transferring a model from one crop species to another, more so from one set of soils to output Addiscott and Mirza, 1998 in spite of many non-linearities in the model. Mesoscopically, another, gave an insight into structural changes needed for modelling winter rye. This crop is these simulations could reproduce a plausible range of yields Butterfield et al., 1997 and an usually well adapted to grow on soils with low PAW. Compared to wheat, it uses 20–30 less intra-regional yield variation similar to that observed in practice Hay et al., 1986. water per unit dry matter produced, and a small and continuous water deficit may even enhance Scaling up the small scale soil map clearly demanded recognition of two soil types with drought resistance and give higher yields Bushuk, 1976. Adaptation or variable response of the different soil properties for modelling crop yields Table 6. Down-scaling the information from the rootshoot ratio under varying climatic conditions has not been included in this model so far. Long- large-scale map, one could not assign the soil type distribution without prior knowledge of the catch- term field trials have shown that apart from water shortage in spring, there is no effect of water deficit ment character. The pixel resolution of the EU soil map 0.5 ° ~2800 km 2 created difficulties in on the yield of winter rye Ro¨mer, 1988. It seems, therefore, that the generalized water stress function locating the validation site 6 km 2 within either of these pixels. The final decision about the impor- derived earlier for wheat Groot, 1987 needs to be modified and seasonally weighted for rye. tant soil properties then becomes an ‘either … or’ decision, which may lead to a difference of 100 As 10 of the simulated yield variation could be attributed to variable N mineralization, there in relevant soil properties Richter and Addiscott, 1998. As with simulation at the national scale was some N stress with respect to potential yield, in spite of ample N fertilizer application. However, Butterfield et al., 1997, knowledge of soil-type distribution within a region became important for the model overestimated uptake rate and did not exert sufficient N stress to reduce dry matter the model output at the catchment scale. Scaling up of the output is a concession to the production. Correct description of the uptake of mineral N from the soil remains a key issue. The unknown deviation of modelling results from observations in environments different from that root system of winter rye is very dense Dittmer, 1937; Ellen, 1993, but little quantitative knowl- of the site of model calibration. Basically, ‘down- scaling’ yields from experimental sites to farmers edge exists about its soil–root interaction with respect to nutrient uptake efficiency in physically fields is justified for two reasons only: 1 sub- optimal management inhibiting ‘obtainable’ yields different environments Troughton, 1962; Goss 250 G.M. Richter European Journal of Agronomy 11 1999 239–253 et al., 1993. High residual nitrate levels after ature during vegetative growth EC 28–61 were much higher than in other years Table 1, and harvest in sandy soils Richter et al., 1998 support the overestimation of yield was also favoured by the suggestion that the actual N uptake is smaller a long grain filling period. The underestimated in coarse, than in fine, textured soils Stark, 1994 yields of 1992, partly due to water shortage and in turn limits potential growth. Table 4, could also be explained by short periods The simulated N returned with the residues for grain filling and pre-anthesis growth Table 7. 50 kgha seems sufficient to equilibrate organic Overall, plant development and dry matter pro- matter loss under row crops and meet the require- duction constitute a complex interactive system, ments for sustainability of the cropping system. with more than 20 parameters related to assimila- The mean N balance in this region was based on tion and assimilate partitioning. Such models need an average N export of 65 kg Nha with an average long-term calibration, even though their parameter winter rye yield of 4.5 tha Kleeberg et al., 1993. sets could be diminished by parsimonious struc- This corresponded to less than 1.5 N in the rye tural changes. grain, which, according to Ruhrstickstoff 1993, These results suggest two structural changes of is far below average. From the lower and upper the model. First, model robustness should be limits of observed yields and N concentrations enhanced by introducing sink limitations. A preset 4.6–6 tha with 15–25 kg Nt, the N export maximum leaf area will prevent excessive pro- would average 100 kg Nha range 59– duction of biomass as in 1990. Although justified 128 kg Nha. This is slightly less than the mod- by diverse plant architecture, this procedure may elled value 112 ±40 kg Nha. The underesti- cause problems in different environments Landau mated N min contents at harvest again confirm the et al., 1998. A more mechanistic approach is the need for model refinement with regard to N introduction of a leaf death rate dependent on uptake. temperature and light competition. Likewise, the However, there is a justified call for model number of flowers and grains per ear can be limited parsimony Webb et al., 1997; Landau et al., and reduced by adverse conditions during flower- 1998. Simplifications of the crop growth model ing Bushuk, 1976. Second, transpiration was include the weather-driven interaction of plant obviously not a yield limiting factor for winter rye, development and the production and partitioning and a weighting factor for water stress at different of dry matter. There are many combinations of phases of crop development should be introduced the three most important environmental variables to account for the crop’s sensitivity to this. during the three major growth phases, even if one expresses climate only as low, medium or high for rain, radiation and temperature. In 19891990,

5. Conclusions